A NNALES DE L ’I. H. P., SECTION A
W. J. C OCKE
A maximum entropy principle in general relativity and the stability of fluid spheres
Annales de l’I. H. P., section A, tome 2, n
o4 (1965), p. 283-306
<http://www.numdam.org/item?id=AIHPA_1965__2_4_283_0>
© Gauthier-Villars, 1965, tous droits réservés.
L’accès aux archives de la revue « Annales de l’I. H. P., section A » implique l’accord avec les conditions générales d’utilisation (http://www.numdam.
org/conditions). Toute utilisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
283
A maximum entropy principle
in general relativity
and the stability of fluid spheres (1)
W.
J. COCKE
(Cornell University (N. Y)
et Institut HenriPoincaré, Paris)
Ann. Inst. Henri Poincaré, Vol. II, no 4, 1965,
Section A :
Physique théorique.
SUMMARY. - The
principle
that the totalentropy
be a maximum isapplied
to
gravitational systems
of aperfect,
relativistic fluid in adiabatic motion.An
expression
for theentropy density
of the fluid is derived as anexplicit
function of the energy
density,
and the conservationequations imply
thatthis
expression
satisfies the propercontinuity equation.
Theentropy principle
is shown toyield
theequation
ofhydrostatic equilibrium
and astability
criterionequivalent
to one derived fromdynamics. Application
to features of
equilibrium
models is studied.SOMMAIRE. - Le
principe d’entropie
maximum estapplique
ausystème gravitationnel :
fluide relativisteparfait
en mouvementadiabatique.
On trouve une
expression
de la densited’entropie,
fonctionexplicite
de ladensite
d’energie.
Lesequations
de conservation entrainent que cette fonc- tion satisfaitidentiquement
a uneequation
de continuité. On deduit duprincipe d’entropie
maximum1’equation d’équilibre hydrostatique,
et uncritère de stabilité
equivalent
a celuiqui
découle de ladynamique.
Desapplications
aux modèlesd’équilibre
sont etudiees.(1)
This paper is a revised version of a thesis presented to theFaculty
of theGraduate School of Cornell
University,
Ithaca, New York, U. S. A., inpartial
fulfillment of the
requirements
for the degree of Doctor ofPhilosophy.
The author is
presently
the holder of a PostdoctoralFellowship
of the National Science Foundation.ANN. INST. POINCARÉ, A-II-4 19
284 W. J. COCKE
INTRODUCTION
This paper deals
chiefly
withsetting
up a maximum entropyprinciple
for
gravitationally
boundspheres
ofperfect fluid,
treated via Einstein’sgeneral theory
ofrelativity.
We will derive anexpression
for the totalentropy
of such a fluidsphere
in terms of the pressure, energydensity,
and
gravitational potentials
andrequire
that thisexpression
be a maximumwith
respect
to smallperturbations conserving
the total energy of the fluid.We will show that this
requirement implies
both the relativisticequation
of
hydrostatic equilibrium
and a criterion for thestability
of staticconfigu-
rations of fluid
satisfying
thisequation.
Considerable interest will lie in the
stability criterion,
and it will be shown that it isequivalent
to onepreviously
obtainedby
S. Chandrasekhar fromdynamical
considerations. We will also use it to elucidate certain features of actualequilibrium
models. Inparticular,
we will consider how unstableconfigurations
are related to the location of maxima and minima ofplots
ofequilibrium
mass curves,parametrized
as functions of the central pressure ordensity.
Throughout
the paper we consider all motions to beadiabatic,
so thatwe
neglect
heatflow,
but we make no otherassumption
as to the form of theequation
of state,except
that it be a function of the usualthermodynamic
variables. We will limit our discussion to
stability against spherically symmetric perturbations, although
the formalism could be extended to include othertypes.
In
general
there are two methods oftesting
thestability
of a static confi-guration
offluid,
both of which have beenapplied
to the Newtonian case.We will now
briefly
discuss these two methods and indicate the connection between them.The first method is known as the «
dynamical
method » and has beenapplied
to both Newtonian and relativistic fluid mechanics[1, 2].
Itinvolves
assuming
that thequantities 03C8j characterizing
the state of the fluidsystem deviate,
as functions ofpositions
andtime, only
veryslightly
from.a set of
given
staticquantities
One may then take thetime-dependent equations
of motion to be linearequations
with the small differences~ 2014
==~
asdependent
variables.If the motions are assumed to be
adiabatic,
and hencereversible,
thetime-dependance
can beseparated
outby assuming
a functionaldependence
on r, t of the form
~y
= cos at, where a is a constant. Theequations
285
A MAXIMUM ENTROPY PRINCIPLE
for
~~~(r)
can then be transformed into a Sturm-Liouvilleeigenvalue
pro- blem, with a2 aseigenvalue.
See inparticular
section I of this paper, wherewe
present
a sketch of thedynamical
method ingeneral relativity.
In
general
the admissibleeigenvalues
a2 form adiscrete, countably
infiniteset, and
stability
is testedby ascertaining
whether or not all the a2’s arepositive.
If not, then some of the a’s areimaginary,
and excitation of these modes will result in anexponential time-dependence
instead of an oscil-latory
one. This will carry theconfiguration
so far away from thegiven
static state that the assumed linear
approximation
will become invalid.A static
configuration
is then stable withrespect
to smallperturbations only
if all theeigenvalues s2
associated with it arepositive.
The second method is known as the « energy method » and has so far
only
beenapplied
in Newtoniantheory [3, 4, 5].
It states thefollowing:
if the total conserved energy E of a mass of fluid can be written as a sum
E = T +
U,
where T is apositive
definite kinetic energy and U is apoten-
tial energy which does notdepend
on thehydrodynamic velocities,
then aconfiguration
is in stableequilibrium only
if U is a minimum withrespect
to infinitesimal variations of the
quantities ~j
which conserve the totalamount
of
material(in
Newtoniantheory,
the totalmass).
That this isnecessary for the
stability
of a static distribution of material isevident,
since otherwise any
lessening
of thepotential
energy Uby
a smallpertur-
bation would release kinetic energy andprovide
furtherimpetus
to thedisplacement.
In terms of the calculus of
variations, requiring
U to be a minimum meansthat we must have 3D = 0 and ö2U >
0,
where ~U is the first variation and ö2U is the second variation. au = 0implies
an Eulerequation,
whichturns out to be
simply
thehydrostatic equilibrium equation,
and a2D > 0 is then astability
criterion.The connection between the
dynamical
method and the energy method is made as follows : it turns out[6]
that a necessary and sufficient condition for 82U > 0 for non-zero variations is that theeigenvalues
fL of aparticular eigenvalue problem
all bepositive.
Thiseigenvalue problem
is ingeneral
not
quite
the same as the one associated with thedynamical
method(as
discussed
above),
but one should be able to show that allthe , ’s
arepositive
if and
only
if all the a2’s arepositive.
This would show that the two stabi-lity
criteria are in fact the same. This has been done in the Newtoniancase
[5, 6].
In
general relativity, however,
the formulation of the energy method iscomplicated by
theapparent
lack of aquantity characterizing
the « total286 W. J. COCKE
amount of material ». In Newtonian mechanics this
quantity
issimply
the total mass of the
fluid,
but inrelativity
the fundamentalequivalence
of mass and energy
implies
that one could alter the total mass of abody
without
actually removing
oradding
any « material».Many
authorshave used the
concept
of a totalparticle
number or so-called « proper mass », but this usage seemsforeign
tohydrodynamics,
which is atheory
of conti-nuous fields
[7].
We may,
however,
turn to thethermodynamic aspect
of theproblem
and
investigate
therequirement
that the totalentropy
of thesystem
be amaximum for
stability.
It is thisprinciple which,
as statedabove,
we willdevelope
in this paper, and with which we propose toreplace
the minimumpotential
energy method.Since we consider
only
adiabaticmotions,
ourexpression
for the totalentropy
S will turn out to be a constant of the motion. We will derivean
expression
for theentropy density
in terms ofquantities entering directly
into Einstein’s field
equations,
and thenverify
from the- fieldequations
that it satisfies the proper
continuity equation.
Anexpression
for the conser-ved total
entropy
S will then follow. This willgive
us a non-trivialintegral
of Einstein’s field
equations
forgeneral
adiabatic fluid motions.We will next show that the statement 8S =
0,
taken relative to variations which conserve the total energy,implies
the relativisticequation
ofhydro-
static
equilibrium.
The additional maximumrequirement
82S 0 willthen be shown to lead to the same
stability
criterion obtainedby
Chan-drasekhar from the
dynamical
method. In the lastsection,
the maximum entropyprinciple
will then beapplied
to some otherinteresting
ques- tions.At this
point
wemight
say a few wordsregarding
theassumption
thatno heat flow occurs. It is of course evident that this
postulate
is neverexactly
satisfied in nature, but we mayregard
it asapproximately
true overappropriately
short intervals of proper time relative to an observermoving
with the fluid. In situations
involving
very stronggravitational fields,
such as in agravitational collapse problem,
thisapproximation
wouldappear to be a propos, in as much as proper times inside the fluid would be much slowed down
compared
to the proper times of external observers.However,
we will not hereinvestigate
thevalidity
of ourassumptions,
but will content ourselves with
exploring
their consequences. It would beinteresting
to see how the methods of this papermight
be extended to include the flow of heat and radiationcoupled
with the matter fields.287
A MAXIMUM ENTROPY PRINCIPLE
We use the
following
notation and conventions. Theexpression
forthe interval
length
will be written aswith Greek indices
taking
on the values0, 1, 2,
3. Latin indices will be understood to assume the values1, 2, 3.
Thus the fluid4-velocity
satisfies = 1.
We choose units so that the
velocity
oflight
c = 1 and the Newtoniangravitational
constant G = 1. Thus we will write the fieldequations
aswhere is the Einstein tensor, which satisfies the conservation
identity
V
03B1S 03B1
=0, ~
«denoting
theoperation
of covariantdifferentiation,
andis the energy-momentum tensor.
The
energy-momentum
tensor of theperfect
fluidwithout
heat flow isthen
expressed
in these units aswhere e is the total proper energy
density, p
is the pressure, and is thehydrodynamic
flow4-velocity.
. 1. - THE DYNAMICAL METHOD
IN GENERAL RELATIVITY
To
bring
into focus some of the ideas of theintroduction,
we herepresent
a sketch of the
dynamical
method oftesting
thestability
of fluidspheres
ingeneral relativity.
We have mentioned that we willonly
considerstability against spherically symmetrical perturbations,
and since any staticconfigu-
ration of fluid will exhibit such symmetry, we way
conveniently
take thisas a
starting point.
It will be useful toemploy spherical polar
coordinates and to define the radiusparameter
rby using
the metric formv and x
being
real functions of r and t.Einstein’s field
equations Sf
= - thengive [8], using
aprime
todenote
()-/o-r, for l.L
= v =1,
288 W. J. COCKE
and
for y = v = 0,
where we have used = 1.
Integration
ofEq. (1-3) yields
For y
=1, v
=0, we
obtainwhere the dot indicates The
equation for ,
= v = 2 wouldprovide
a fourth
equation (There
are fourindependent unknowns,
which may be taken to be e,ul,
x, and v, pbeing given
as a function of the other variables.See section
II). However,
it is more convenient to usethe l-L
= 1component
of the conservationequation V
= 0. We shall not write it out infull,
but rather refer the reader to the literature
[9].
We now denote
where so, po,
a~,
vo are solutions of the staticequations
or
and of the static
part
of the conservationequation
=0,
which readsEquation (1-6)
is theequation
ofhydrostatic equilibrium,
which we willderive
again
from the maximum entropyprinciple
in section III. Itis,
of course, theanalogue
of the Newtonianequation dp/dr
= -pdp/dr,
p
being
the massdensity and p
thegravitational potential.
Following
Chandrasekhar[7],
we assume that8e, 8p, 8À, 8v,
and u1 are289
A ENTROPY PRINCIPLE
small
quantities
defined on the interval 0 rRo,
whereRo
is the boun-dary
radius of the staticdistribution ;
and we write to first orderof smallness
Letting ~(r, t )
be the «Lagrangian displacement
)) of an element of the fluid from an initialposition,
we have to first orderand
integrating Eq. ( 1-5),
weget, with U4
"’eVo/2,
From V
a Tla
=0,
to firstorder,
one may obtain the relationSince this
equation
is linear in thedependent
variables and containsonly
the second time
derivative,
we mayseparate
variablesby assuming
a time-dependence
of the form cos at, abeing
a constant,getting
where we now consider
8 ~, 8v, 8p, oe, and ~
to be functions of ronly.
In the next section we show that for adiabatic motions we may
write p
in the
form p
=p(e, x),
where the variable x indexes the fluid element itself and follows its motion. See also reference[1].
Then we have inour case
and we may use this relation to substitute for
ap
inEq. (1-7),
anduse
Eq. (1-2") (1-4")
and( 1-5 ")
to eliminate othervariables, finally writing
Eq. (1-7) with ç
asdependant
variable290 W. J. COCKE
where one can check to see that P ~ 0.
Multiplication by
thequantity q(r)
= exp allows us to write the formwhere
q(r), v(r)
> 0. Theboundary
conditionson ç
arearranged
so thatthe pressure vanishes to first order on the new
boundary
radiusRo
+ç(Ro).
This condition reads
which the reader may
easily
write himself in termsof ç.
At r = 0 we must have ofcourse ç
= 0.Equations (1-8)
and(1-9)
constitute a Sturm-Liouvilleeigenvalue problem
The
problem
is of thesingular
typesince q
and v may be shown to tend to 0 at both ends of the interval 0 rRo. However,
one may show[2]
that in
spite
of thesingular
character of theproblem,
there exists an infinite discrete set ofeigenvalues { 6~ },
of which there is a least memberao-
It is
easily
shown[2, 10]
that we may solveEq. (1-8)
for a2 asthus
defining
the functionals3)[~], JC[~].
Theboundary
terms vanishbecause
q(0)
=q(Ro)
= 0.According
to theorems on the extremum pro-perties
ofeigenvalues [2, 10]
the lowesteigenvalue 6o
isgiven by
where 03C6
ranges over all functions withbounded, piece-wise
continuousp’
which
satisfy
theboundary
conditionsfor ç
and do not vanishidentically.
The condition
for
stability
was firstapplied
to thisproblem by
S. Chandrasekhar[1].
Note that
since q, v > 0,
if t ~ 0 for all 0 ~ rRo,
thenG~
> 0 would followimmediately,
and any suchconfiguration
would bedynamically
stable.
Having
studied theapplication
of thedynamical
method ingeneral
rela-tivity,
we nowproceed
todevelope
ourreplacement
of the energymethod,
the maximum entropyprinciple.
291
A ENTROPY PRINCIPLE
2. - THE TOTAL ENTROPY
In the introduction we stated the need for
expressing
the totalentropy
S of the fluidsphere
in terms of variablesalready entering
into the field equa- tions. Toaccomplish
thisend,
we now construct a relation between the proper entropydensity s
per unit volume and the pressure p and energydensity
e.First let V be the volume of a very small element of the fluid. Since we
consider all motions to be
adiabatic,
the totalentropy
S = sV of the ele- ment will be constant over suchmotions,
and we may write thethermodyna-
mic
identity
for the element as dE +pdV
= TdS= 0;
or, since E =sV,
Now since a
change
in volume is theonly
means ofchanging
the thermo-dynamic
state of the fluidelement,
heat flowhaving
beenoutlawed,
itfollows that any
change
in state will beaccompanied by
a non-zerochange
in e,
except
for elements inwhich p
and e are both zerosimultaneously.
Thus, except
for such «empty » elements,
we may say all thethermodynamic
variables may be written as functions of the energy
density
e,including
of course the pressure.
However,
theequation
of state p =f(e)
will then ingeneral
be a different function of e for different fluid elements.Let us now consider the use of a
co-moving
coordinatesystem
in which thespatial components
uj(~’ = 1, 2, 3)
of the fluid4-velocity
vanishthroughout
thesystem
for all time. Such coordinates arepresumably always possible
sincethey
may be definedby simply attaching
thespatial
reference
system
to the identifiablepoints
of the fluid.By
virtue of the above arguments we may now write the pressure as a function of e and of thespatial
coordinates xj of theco-moving
frame:p =
p(e,
the xjserving
to index the differentpoints
of thefluid ;
i. e., the different fluid elements. It must be undertood that the functionp(e, xi)
is to be derived from
purely thermodynamic
considerations afterspecifying
the initial pressure and energy
density
distributions at a time x° =But since the total entropy of the fluid element S = sV is
conserved,
wemay form d S =
d(sV)
= sdV + Vds =0,
or -dV/V
=ds/s. Hence,
fromabove,
292 W. J. COCKE
Integration
with xi = constantyields
or
where so and so are functions of xj such that
s(eo)
= so.We have thus
expressed
theentropy density s
as a definite function of the energydensity e
and thespatial
co ordinates xj of theco-moving
frame jc~.Let it be
emphasized again
that thisexpression
is validonly
for adiabaticchanges
of state.Note that of it is
possible
to define a conservedparticle
number for thesystem such that n is the proper
particle
numberdensity,
the totalparticle
number N = nV of a small fluid element would also be
conserved;
andlikewise, - d V/V
=dn/n.
Hence we wouldsimilarly
have°
and thus n and s would be
proportional
to the same function of s for fixed xj.As an
example
let us find anexpression
for theentropy density
of a per- fect non-relativistic Boltzmann gas with constantspecific
heat. We may write theequation
of state for an element of such a gas aspV
=NkT,
or p = nkT. For adiabatic
compression
orexpansion
we furtherhave,
with y as the constant ratio of
specific
heats perparticle
y =cp/cv [11], pVY
= constant or p = constant nY, or n = where no and paare functions of xj. Then the energy
density
isgiven by
where mo is the rest mass per
particle
and Eo is anarbitrary
constant. There-fore
which definies p =
p(e, xj)
as animplicit
function. Thenwhere we have used cp - Cy = k. It then follows that
We have thus verified the
proportionality
of sand n.293
A MAXIMUM ENTROPY PRINCIPLE
The total
entropy
of the fluidelement
may be written[11]
where ~
is the chemical constant of the gas. HenceThis verifies the
expression
for s derived above fromEq. (2-1).
We must now show that our
expression
for s satisfies the proper conser- vationequation
= 0. Let us write the scalar in terms ofour
co-moving
frame in whichonly
u° isnon-vanishing :
where indicates Now we have from
Eq. (2-1)
thatso that
and hence
But this is a tensor
equation,
and hence is true notonly
in theco-moving
frame but in anysystem
of coordinates whatever. Now the fieldequations imply
thatLet us further construct the scalar
product
um v =0, using
the rela-tion = 1 and its
consequent
= 0 :Comparison
withEq. (2-3)
then shows that infact,
if the fieldequations
are = 0.
294 W. J. COCKE
We may use this relation to show that the total proper
entropy
of the fluidsystem
is a conservedquantity.
Let t be a time-like coordinate and x, y, z bespace-like
coordinates(not necessarily
Galilean atinfinity).
Let2p
be thespace-like
surface of the mass offluid,
and let~a
and~b
be the intersections of the time-like surfaces t = a and t= b, respectively,
withthe
space-time
volume of the fluid. Then ifF
be thepart
of03A3F
which lies between~a
and~b,
the surfaces~Q, ~b,
and2p together
form a closed3-space E
inspace-time.
Now let V denote the
space-time
volume of fluid enclosedby E.
Thenfor an
arbitrary
vector fieldA~,
we may write Green’s theorem for this volume and surface as[12]
where n~ is the outward unit normal to
~,
and and aregeneralized
3- and 4-volume elements.
If
= 0 is theequation
ofI:,
then we havewith the
sign
chosen so that n~ is directed outward from E. Inside Vwe can let =
il’ -
gdt dx dy dz
and onEa
and~b
we can set=
Ý -
g~3~ dxdy dz,
where is the determinant of the subtensor gij(i~ .I
=1, 2, 3).
Then if we choose A~ = we will have = = 0 inside
V,
and
consequently, by
Green’stheorem,
Now the
equation
for~a is
= t - a =0,
so that onE.,
for ab,
Similarly,
on1~
Thus it follows
that,
since295
A ENTROPY PRINCIPLE
Further,
we must alsosatisfy
the Lichnerowiczjunction
conditions[13],
from which one may
easily
deduce that on the surface2p,
ua.na. = 0. Thiscan also be shown
by expressing
u~ and ntL inco-moving coordinates,
inwhich the
equation
forEp
will have theform f(xj)
= 0.Thus we obtain
and hence
for all t = a, b.
Or,
where
2~
is the3-space t
= constant. Theintegral
S may be said to repre- sent the totalentropy
in the same sense that conservation ofcharge
in elec-trodynamics
in the form =0,
where p is thecharge density,
leads toan
exactly
similarexpression
for the total conservedcharge.
We now
proceed
to formulate our variationalprinciple
S = maximumwith
respect
to variations which conserve the total energy.3. - THE FIRST VARIATION
We now
require
that forhydrostatic equilibrium
the totalentropy
S be an extremum(maximum)
for all infinitesimal adiabatic variations of the energydensity e
and pressure p which conserve the total energy of the sys- tem. Asstated,
we will restrict ourselves to variations which preserve thespherical symmetry
of thefluid,
and thus we will be able to use the expres- sions for the metric form and for the fieldequations presented
in section I.It should be stated at the outset that in
performing
the variations wewill not consider the variations as
independent
from ~e and8p,
butwill rather let them be
given
via the fieldequations (1-2") ( 1-4")
and(1-5").
We shall not use the
equations T2
= -803C0S22
and =0,
sincethey
themselvesimply
theequation
ofhydrostatic equilibrium. Further,
thevariation
öp
will be linked to ~sby
means of the adiabaticequation
of statep =
p( e:, representing,
as it were, thepoint
oforigin
of the fluid element. Thus we will haveonly
oneindependent
variational function.296 W. J. COCKE
Let us also note that even
though
the variations areadiabatic, they
willnot conserve the total
entropy identically,
since we willalways
consider thespatial velocity
of the material to be zero.In order to carry out our variations we will need an
expression
for thetotal energy. It is
easily
shown that for the metric from(1-1),
a conservedenergy may be written in the forms
[14]
Thus for the static
problem
we take ui = 0 and writeUsing Eq. (2-4),
one may then express the totalentropy,
afterperforming elementary integrations,
asBut with u1 =
dr/ds
=0,
we have fromEq. (1-1)
thatdt/ds
=Hence
However,
intaking
the variations of M andS,
we must allow for the fact that theboundary
radius R will also vary. Thus it behooves us to trans- form theindependent
variable r so that the condition M = constant auto-matically
fixes the endpoints
of the interval ofintegration.
We do thisby transforming
to the variablewhich
represents
thequantity
of energy inside the radius r.We denote
dr/dm by
r(we
trust there will be no confusion with a time.
derivative in this
context)
and finddm/dr
= or = r =and thus
Also we have
297
A ENTROPY PRINCIPLE
We may hence express the
integrand
of S as a function of theindependent
variable m and the
dependent
variable r and its derivative r. Theprinci- ple
~S = 0 will then determine r =r(m)
from the Eulerequation,
and thecondition M = constant will be
equivalent
tofixing
the upper endpoint
of the interval ofintegration
0 m M.However,
careful note must be made of the fact that thedependence
ofthe
entropy density s
=xi)
on thespatial
variables xi is such that xialways
follows the motion of the fluid element. Thus inreality,
we havefor the
spherically symmetric problem s
=s(e,
r -Q, where ~
is the dis-placement
of the element of fluid which finds itself at r afterhaving
beenmoved from the
point
r -ç. ç
is thengiven
in terms of the other varia- blesthrough
the relation( 1-5 ").
Let us denote ~ == r -ç,
so thats =
x).
Our next
step
would be to write the variation of S in terms of the varia- tion or = of thedependent
variable r.However,
we must first beable to
express ~
in terms of Sr. Here is the infinitesimalchange
ofthe radius r =
r(m)
of asphere
of fixed energy content m, and not theLagrangian displacement,
which we havedesignated
as~.
Let us denote the
change
in energy contentm(r)
of asphere
of radius rby
=403C0r0
03B4~r2dr. Thenwriting mo(r)
as theequilibrium
distribution0
function around which the variation is
taken,
we haveby
definitionand hence
to first order.
Thus, using Eq. ( 1-4")
and( 1-5 "),
we obtainWe now write
298 W. J. COCKE
The first variation of S will thus be
expressed
asAccording
to the fundamental lemma of the calculus of variations the condition 3S = 0implies
the Eulerequation
and the condition = 0 since in
general ~r(M) 5~ 0, although
of course8r(0)
= 0.We first
investigate
theboundary
term : We haveand
using Eq. (3-3),
whichgives
= -(403C0r2r2)-1
== -~/r
one obtainseasily Lr - e03BB/2r2sp(p
+e)-I.
Thus if the energy
density
does not vanish at thesurface,
we will have= 0
through
the condition that the pressure on the surface vanish.But suppose e = 0 also. Then we may write
Lr -
anddemand
simply
that remain finite as e - 0. This iscertainly
a very reasonablestipulation,
and one canshow,
forexample,
that forequations
of state of the form e = «p ~--
(see Eq. (2-2)), stays
finite as e2014~0 forpositive
«,p,
y if andonly
if y > 1.However,
one may also show that theproduct
goes to zero as e --~ 0 for all y >0,
so thatLr]M
= 0still
tells us very little about theequation
of state.We now
proceed
to examine the Eulerequation (3-6).
Let us writed(Lr )/drn
=(dr/dm)d(Lr )/dr
= and expressEq. (3-6)
asa total differential
equation
with r asindependent
variable asbefore, again using
aprime
to denoted/dr.
We havewhere
Eq. (3-3)
and(3-4) give
299
A ENTROPY PRINCIPLE
and
Further
But
so that the terms in will cancel. Then after
using Eq. (1-3’)
andcollecting
terms, we may writeEq. (3-6)
aswhich
with the aid ofEq. (1-2’)
may be writtenp’(p
+ +v’/2
=0,
which we
recognize
asEq. (1-6),
theequation
ofhydrostatic equilibrium.
Thus we have obtained
Eq. (1-6) by
means of the maximumentropy principle
and with thehelp
of theequations S~
= -8?rT~ S~
= -and
S~
= - We nowproceed
to examine the second variation ~2S andrequire
that it benegative.
4. - THE SECOND VARIATION AND
EQUIVALENCE
WITH THE DYNAMICAL METHOD
Now that we have found a necessary condition
(the
Eulerequation)
for Sto be an extremum, the
principle
S = maximumrequires
us toinvestigate
82S 0 for infinitesimal variations about the extremum. We will show that this
requirement
isequivalent
to be criteriondeveloped by
Chandra-sekhar
by
means of thedynamical
method. It will be shown in fact that if~o
be the leasteigenvalue satisfying Eq. (1-8),
thenao
> 0 if andonly
if a2s 0 for all non-zero infinitesimal variations
conserving
total energy andsatisfying
the usualcontinuity
andboundary
conditions.We will use
again
theindependent
variable m asdevelopped
in the last section to accommodate variations in theboundary
radius R and tocomply automatically
with M = constantby fixing
the upper endpoint
of the inter- val ofintegration.
ANN. INST. POINCARÉ. A-II-4 20
300 W. J. COCKE
The second variation 82S is
given by [6, 16]
Substitution of ~r
for ~
viaEq. (3-5)
then allows us to write 82S in the formthus
defining
thequadratic
formQ(8r, 03B4r).
A rather
simple
test to be satisfied for S to be a maximum is the so-calledLegendre-Weierstrass
test, which is a necessary condition for 828 0 withrespect
to bothstrong
and weak variations. It is a local test and is writ- ten[16, 17]
where rand
r
are actual extremalvalues, p
is anyphysically
realizable value of r, and 6 is any number between zero and one. Theequality sign
canhold
only
for isolatedpoints.
But
where the
dependence
onr
isthrough
£ =r),
and hence forpositive
£the test is satisfied if >
0,
whichcertainly
holds for any real one-phase
fluid. This also shows that Scertainly
cannot be a minimum.We now
proceed
to formulate a moregeneral
test(related
to the so-called Jacobi test[18])
which includes theLegendre-
Weierstrass test in its formula- tion and which we will show isequivalent
to Chandrasekhar’s criterion.First we demonstrate that a necessary and sufficient condition for ö2S 0 is that all the
eigenvalues ~
of the associated linearproblem (abbreviating
etc.)
must be
positive.
Theboundary
conditions to beapplied
infinding
the are
given
asthe latter of which may be shown to be the same as
Eq. (1-9),
which statesthat the pressure must vanish to first order on the
varying boundary
of thesphere.
301
A MAXIMUM ENTROPY PRINCIPLE
Now let us
multiply Eq. (4-1) by
or andintegrate, getting
Again exploiting
the extremalproperties
ofeigenvalues,
one mayeasily
show that for the smallest
eigenvalue
ll-o,for all
admissible 03C6
notidentically
zerosatisfying
the sameboundary
andcontinuity
conditions as ~r. Thus if yo >0,
then 82S0,
andconversely.
We are now in a
position
to compare our condition 82S 0 with Chan- drasekhar’scriterion,
which as we have said results fromlinearizing
thetime-dependent equations
of motion.Substituting ~
for Sr inEq. (4-1)
with the use of
Eq. (3-5),
one may, afterchanging
theindependent
variableback to r and
performing
very tediousmanipulations,
obtain anequation
of the form
where q, u > 0.
Comparison
withEq. (1-8),
theeigenvalue equation
for a2 in Chandrasekhar’s
criterion,
would show thatq(r)
andt(r)
are thesame
functions
in bothequations,
but that ingeneral u(r) ~ v(r).
Thesame
boundary
conditions of course hold for bothequations.
We will now prove that all the
eigenvalues
l1. ofEq. (4-2)
arepositive
if and
only
if all the a2’s inEq. (1-8)
arepositive.
Asbefore,
we writefor the lowest
eigenvalues
f1.o and~o, using q(0)
=q(R)
=0,
and
the functional
being
of course the same in bothequations.
Now onecan show
[2]
that there exists an admissible cpo such that302 w. J. cocKE
Now suppose 0.
0,
and henceTherefore 0
implies ao 0,
and one maysimilarly
show that0
implies
0. Hence all the[.L’S
arepositive
if andonly
if allthe a2’s are
positive,
and hence 82S 0 if andonly
ifcr:
> 0.This shows the
equivalence
of the twostability
criteria. We now passon to consider
applications
of our formalism to certain features of actual models.5. - APPLICATION TO FEATURES
OF ACTUAL MODELS
Actual
equilibrium
models for avariety
ofequations
of state have beenfairly thoroughly investigated,
and thestability
of some of these models hasbeen tested with Chandrasekhar’s criterion
[19].
Since the results whichwe will discuss
[19, 20]
were obtained forequations
of state of the formp =
peE)
with nodependence
on aco-moving
variable x, we will use p =p(e) throughout
this section as asimplifying assumption.
One
aspect
of the results obtained takes the form of curves of total energy(mass)
of the models as a function of aparameter
such as the central pres-sure p~. We write M =
MoCPc).
One may show thatonly
one suchfunction
Mo(pc)
is obtainable for agiven equation
of strate.A
stability analysis
for agiven pee)
then shows which values of pc are associated with stableconfigurations
and which with unstable ones. One feature thatusually
appears is that for agiven equation
of state there isa
point ~
such that for 0 ~ pcp°
theconfigurations
arestable,
whilefor pc >
p~
theconfigurations
are unstable. We call thepoints p~
« tran-sition
points )). Further,
the transitionpoints po
are known to be « masspeaks
», or maxima of the functionsMo(pc). However,
it appears that not all masspeaks
are such transitionpoints.
We will prove aplausible
suffi-cient condition for a mass
peak
to be apoint
of either neutral or unstableequilibrium.
This willapply
as well to minima on the mass curves.Oppen-
heimer and Volkoff
[20]
havepresented
arguments which lead one to the conclusion that all maxima and minima must be suchstability-instability
transition
points.
Calculationsby
Misner andZapolsky [19]
have shownthat this is not the case, and we will
explain why Oppenheimer
and Vol-koff’s